Leveraging neighborhood session information with dual attentive neural network for session-based recommendation

In the context of user uncertainty and limited information, predicting user preference is a challenging work in many online services, e.g., e-commerce and media streaming. Recent advances in session-based recommendation mostly focus on mining more available information within the current session. However, those methods ignored the sessions with similar context for the current session, which contains rich collaborative information. Therefore, in this study, we proposed a novel Leveraging Neighborhood Session Information with Dual Attentive Neural Network (LNIDA) for session-based recommendation. Specifically, LNIDA contains two main components, i.e., Current Session Encoder (CSE) and Neighborhood Session Encoder (NSE). The CSE module exploits an item-level attention mechanism to model user's own information in the current session, and the NSE module further captures neighborhood collaborative information via a session-level attention. Then, a simple co-attention fusion mechanism is used to dynamically combine information from the CSE and NSE. Finally, to verify the performance of LNIDA, we conduct extensive experiments on three benchmark datasets, YOOCHOOSE and DIGINETICA, and the experiment results clearly show the effectiveness of LNIDA. Furthermore, we find out that LNIDA can improve performance when modeling the current session information and the neighborhood session information simultaneously. (c) 2021 Elsevier B.V. All rights reserved.

Automated summary

In this study, a novel session-based recommendation method LNIDA is proposed to improve recommendation effectiveness by utilizing neighborhood session information and a dual attentive neural network. Experimental results on three benchmark datasets show the effectiveness of LNIDA in modeling both current session information and neighborhood session information simultaneously.